基于LSTM的一次风机在线监测及故障诊断研究

LSTM-Based Research on Online Monitoring and Fault Diagnosis of Primary Air Fan

鉴于一次风机的运行状态对于火电机组安全运行具有重要意义,文章提出了一种一次风机状态监测与故障诊断的数据分析方法。采集M电厂一次风机正常状态运行数据,进行数据预处理,计算各参数皮尔逊相关系数,根据结果选取相关性大的参数对y方向轴瓦振动、自由端轴承温度、x方向轴瓦振动进行回归。使用最小二乘支持向量机(LS-SVM)模型、长短记忆神经网络(LSTM)模型对x、y方向轴瓦振动、自由端轴承温度进行回归,计算回归结果均方根误差。对正常状态与异常状态三参数进行LSTM回归,计算回归值自适应阈值,仅异常状态x轴瓦振动实际值超过阈值上下限,认为LSTM模型可适用于一次风机的状态分析与故障诊断。

In view of the great significance of the operation status of primary fan for the safe operation of thermal power units, this paper proposes a data analysis method of primary fan state monitoring and fault diagnosis. Collect the normal operation data of the primary fan in M power plant, conduct data preprocessing, calculate the Pearson correlation coefficient of each parameter and, according to the results, select the relevant parameters for regression analysis of y shaft shoe vibration, free end bearing temperature and x shaft shoe vibration. LS-SVM model and LSTM model are used to regress the x,y shaft shoe vibration and free end bearing temperature to calculate the root mean square error of regression. The LSTM regression of the three parameters in normal state and abnormal state is conducted to calculate the adaptive threshold of regression value. Only the actual value of the abnormal state of shaft shoe vibration in x direction exceeds the upper and lower limits of the threshold value. It is considered that the LSTM model is applicable to the state analysis and fault diagnosis of the primary fan.